Present manual discovery and troubleshooting methods are insufficient. In this thesis a fundamentally new approach is proposed. It's based on discrete event analysis application on network signaling as a voice call can be described as sequence of messages from the network's perspective. This solution could not just significantly reduce network failure management efforts, but could also offer additional insight of the nature of malfunctions.
In order to design and implement a proof of concept library and demo application, related literature was surveyed in depth and appropriate (dis)similarity metrics and clustering methods were chosen for both off-line and online analysis. Challenges and conclusions regarding the design and implementation processes are highlighted.
Constant reviewing and comparison is following every worked out and used metrics and clustering solutions from various perspectives. These findings and several months of experience gained from testing on real file data is being discussed. Final evaluation shows that our proposed approach performs well and our proof of concept fulfilled all the objectives set by being more then capable of providing near real valuable time network failure insight.